fix: CheckLLMResponseStream QPS is too high

Author: ylsssssss

volcenginesdkwafruntime/__init__.py

@@ -15,5 +15,6 @@
 from volcenginesdkwaf import *
 from volcenginesdkwafruntime.api.waf_runtime_api import  WAFRuntimeApi
 from volcenginesdkwafruntime.models.llm_stream_session import LLMStreamSession
-
-__all__ = ["WAFRuntimeApi", "LLMStreamSession"]
+from volcenginesdkwafruntime.models.llm_stream_session import  LLM_STREAM_SEND_EXPONENT
+from volcenginesdkwafruntime.models.llm_stream_session import  LLM_STREAM_SEND_BASE_WINDOW
+__all__ = ["WAFRuntimeApi", "LLMStreamSession" , "LLM_STREAM_SEND_EXPONENT" , "LLM_STREAM_SEND_BASE_WINDOW"]

volcenginesdkwafruntime/api/waf_runtime_api.py

@@ -1,6 +1,6 @@
 
 from volcenginesdkwaf import WAFApi, CheckLLMResponseStreamRequest
-from volcenginesdkwafruntime.models.llm_stream_session import LLMStreamSession
+from volcenginesdkwafruntime.models.llm_stream_session import LLMStreamSession,LLM_STREAM_SEND_EXPONENT,LLM_STREAM_SEND_BASE_WINDOW
 
 global_llm_send_len = 10
 
@@ -52,13 +52,15 @@ def check_llm_response_stream(
 
             # 重置流缓冲区和发送长度
             session.set_stream_send_len(0)
+            session.CurrentSendWindow = session.CurrentSendWindow * LLM_STREAM_SEND_EXPONENT
 
             return response
 
         # 3. 处理 use_stream 为其他值的情况（累计长度，超过阈值才发送）
         else:
             # 如果未发送长度超过 10 个字符，调用 API
-            if session.get_stream_send_len() > global_llm_send_len:
+            if session.get_stream_send_len() >= session.CurrentSendWindow:
+                session.CurrentSendWindow = session.CurrentSendWindow * LLM_STREAM_SEND_EXPONENT
                 # 准备请求体，使用 session 中的完整流内容
                 body.content = session.get_stream_buf()
                 body.msg_id = session.get_msg_id()

volcenginesdkwafruntime/models/llm_stream_session.py

@@ -2,10 +2,11 @@
 
 from volcenginesdkwaf import CheckLLMResponseStreamResponse
 
+LLM_STREAM_SEND_BASE_WINDOW = 10
+LLM_STREAM_SEND_EXPONENT = 2
 
-class LLMStreamSession:
-    """对应 Java 中的 LLMStreamSession 类"""
 
+class LLMStreamSession:
     def __init__(
             self,
             stream_buf: str = "",
@@ -17,6 +18,7 @@ def __init__(
         self.stream_send_len = stream_send_len
         self.msg_id = msg_id
         self.default_body = default_body
+        self.CurrentSendWindow = LLM_STREAM_SEND_BASE_WINDOW
 
     def get_stream_buf(self) -> str:
         return self.stream_buf